Exploring memory-related network via dorsal hippocampus suppression

Memory is a complex mental process in the mammalian brain. Extensive research has been carried out to identify individual brain regions involved in memory, such as the hippocampus, thalamus, and prefrontal cortex (Danet et al., 2015; Euston et al., 2012; Hainmueller & Bartos, 2020; Murray et al., 2007). In addition, successful memory processes require coordinated large-scale communication between these key brain regions (Aggleton et al., 2010; Opalka & Wang, 2020; Rajasethupathy et al., 2015). However, how brain regions in the memory-related network orchestrate their activity for memory process remains elusive.

To begin unraveling the relationship between brain network activity and memory performance, one research strategy involves the intentional perturbation of a brain node in the memory network, along with subsequent measurement of the corresponding changes in network activity/connectivity that covary with perturbation-induced memory changes at the behavioral level. For this purpose, the dorsal hippocampus (dHP) is an ideal candidate region because of its pivotal role in multiple memory processes: encoding, storage, and retrieval (Hainmueller & Bartos, 2020; Liu et al., 2012; Restivo et al., 2015; Park et al., 2016). Additionally, spatial and object information are encoded in the dHP as neural ensembles (Danielson et al., 2016; Hainmueller & Bartos, 2018). Anatomically, dHP is composed of dorsal regions of the dentate gyrus (DG), CA3, and CA1. It receives input from the entorhinal cortex (EC) via the perforant pathway, which is in turn connected with the rest of the cortex (Sewards & Sewards, 2003; Witter et al., 2017). The dHP, through complex neural pathways mediated by the fornix, communicates with various brain structures such as the thalamus and mammillary complex, as well as the cortex (Aggleton et al., 2022). These interactions facilitate the hippocampus’ role in memory and spatial navigation. Given this extensive connectivity structure that might be implicated in memory process, a perturbation to the dHP could impart a widespread effect to other regions in the memory-related network.

In order to provide a targeted, local perturbation of the dHP, here we employed a chemogenetic approach in rats. In combination with neuroimaging techniques and behavioral assessment, we aimed to establish a relationship between memory network dynamics and memory performance. More specifically, neural activity of the dHP was suppressed using the designer receptors exclusively activated by designer drugs (DREADD) technique (Armbruster et al., 2007). Resting-state functional magnetic resonance imaging (rsfMRI) experiments were conducted to measure brain-wide changes in the spontaneous blood oxygen level–dependent (BOLD) signal and resting-state functional connectivity (RSFC) following acute DREADD-induced suppression. On separate days from imaging, spatial memory behavioral testing (Y-maze) (Conrad et al., 1996; Jung et al., 2008) also took place after DREADD-induced suppression to investigate relationships between neural network changes and the corresponding changes in memory performance. rsfMRI data was acquired without the use of anesthesia, permitting a linkage between the imaging and behavioral data (Beloate & Zhang, 2022; Dopfel & Zhang, 2018; Tu et al., 2021). Moreover, as DREADD actuators can interact with multiple and diverse endogenous receptors (Jendryka et al., 2019; Martinez et al., 2019; Nagai et al., 2020), we separately utilized two DREADD actuators (CLZ and CNO) across all experiments to control for possible off-target effects. Our results show that dHP suppression causes brain-wide alterations of RSFC that are correlated with spatial memory performance. This data offers new insights into structures of the memory network that may be necessary for proper memory function.

Data were collected from 36 male Long-Evans rats weighing between 300 and 400 g. Rats were singly housed postsurgeries with food and water provided ad libitum. The housing room was maintained at constant room temperature and adhered to a 12-hour light and 12-hour dark cycle. All experiments in the current study were approved by the Institutional Animal Care and Use Committee at the Pennsylvania State University.

The rat brain was injected with adeno-associated viral vectors (AAVs) by way of aseptic stereotaxic surgery. Following a brief induction of isoflurane anesthesia, an intramuscular injection of ketamine (40 mg/kg) and xylazine (12 mg/kg) was administered to maintain the anesthetic state during surgery. Buprenorphine ER-LAB (1 mg/kg) was subcutaneously administered to provide postsurgery analgesia. Dexamethasone (0.5 mg/kg) and Enrofloxacin (2.5 mg/kg) were subcutaneously administered as an anti-inflammatory drug and antibiotic, respectively. Bupivacaine (4 mg/kg) was injected under the scalp for local analgesia. Heart rate and SpO2 were continuously monitored using a pulse oximeter (MouseSTAT Jr., Kent Scientific Corporation) during surgery while a heating pad (PhysioSuite, Kent Scientific Corporation) placed underneath the rat maintained the body temperature at ~37°C. Craniotomies were made on the skull for viral injection. For animals in the DREADD groups, AAV8-hSyn-hM4Di-mCherry (1 μL per site, titer ≥ 3 × 10¹² vg/mL, Addgene) was stereotactically injected into the dHP (coordinates: AP −3.5 mm, ML ± 2 mm, DV −2.5 to −2.9 mm). Thirteen rats received unilateral injection in the right dHP (termed unilateral DREADD group hereafter), and 15 rats received bilateral injection in the dHP (termed bilateral DREADD group hereafter). For the sham group (n = 8), AAV8-hSyn-EGFP (1 μL per site, titer ≥ 3 × 10¹² vg/mL, Addgene) was injected bilaterally in the dHP. Bone wax (Ethicon Inc.) was used to prevent bleeding from the burr holes. Animals were given at least 4 weeks for full recovery and sufficient DREADD protein expression (Tu et al., 2021).

Two DREADD actuators were separately used in this study: clozapine (CLZ, Sigma-Aldrich) and clozapine N-oxide (CNO, Sigma-Aldrich). CLZ was dissolved in DMSO and 0.1 N HCl and then diluted with saline to the concentration of 0.1 mg/ml. CNO was dissolved in DMSO and then diluted with saline to the concentration of 1 mg/ml. Control vehicles were saline solution with 0.5% DMSO. Across all experiments, the CLZ (0.1 mg/kg), CNO (1 mg/kg), and saline injections were administered intraperitoneally (IP) while animals were anesthetized with isoflurane vapor. For all behavior tests, isoflurane exposure was brief and immediately discontinued following IP injection. For saline and CLZ, animals were given at least 10 minutes to wake up before behavior tests commenced. In accordance with the literature, at least 30 minutes were given for behavior tests involving CNO injections (Jendryka et al., 2019). For imaging experiments, due to subject set up and prerequisite imaging sequences, at least 30 minutes elapsed between IP injection and rsfMRI data acquisition.

To minimize animals’ stress and motion during awake fMRI imaging, rats were habituated to the restraining and fMRI scanning environment for 7 days (15, 30, 45, 60, 60, 60, and 60 min/day, respectively). Detailed acclimation procedures can be found in our previous publications (Han et al., 2022; Tu et al., 2021). All fMRI data were acquired on a 7T Bruker BioSpec 70/30 scanner interfaced with ParaVision 6.0.1 (Bruker, Billerica, MA). Prior to imaging, rats received IP injections with either saline, CLZ or CNO. T2*-weighted gradient-echo images covering the whole brain were acquired with an echo plana imaging (EPI) sequence using the following parameters: flip angle = 60°, repetition time (TR) = 1,000 ms, echo time (TE) = 15 ms, slice thickness = 1 mm, slice number = 20, field of view (FOV) = 3.2 × 3.2 cm², in-plane matrix size = 64 × 64. 600 volumes were collected for each fMRI scan, and three scans were collected for each fMRI session. A given animal was subjected to at least three fMRI sessions, allowing for each rat to have at least one session with each of the three types of injection solutions (saline, CLZ, or CNO), which were randomly ordered.

The fMRI signal preprocessing used an in-lab developed pipeline detailed in Ma and Zhang (2018) and Liu et al. (2020). The first 10 volumes were removed to ensure a steady state of MR signal. Framewise displacement (FD) was calculated to estimate the motion level of each fMRI volume. Volumes with FD > 0.5 mm were discarded together with their two previous and five following volumes. Volumes with FD in the range of 0.3–0.5 mm were discarded with one previous and three following volumes. Volumes with FD in the range of 0.2–0.3 mm were discarded with their immediate neighboring volumes. Scans that had over 20% of volumes removed were excluded from further analysis. Motion-scrubbed fMRI images were registered to a predefined T2-weighted image template through rigid body transformation. Registered images were motion corrected using SPM12 package, and distortions of the images were corrected using the Advanced Normalization Tools (ANTs) (Avants et al., 2008). Independent component analysis (ICA) (Calhoun et al., 2001) was applied to identify nonneural artifacts. Time courses of noise IC components, motion parameters, and signals from the white matter and cerebrospinal fluid (CSF) were voxel-wise regressed out from the fMRI signal. fMRI data were then spatially smoothed (FWHM = 1 mm). Finally, for RSFC analysis, fMRI data were temporally bandpass (0.01–0.1 Hz) filtered and truncated to the first 480 volumes.

After all imaging experiments, each rat was tested in three locomotion-assessment sessions with either saline, CLZ (0.1 mg/kg), or CNO (1 mg/kg) injection; all rats received all injection types in a randomized fashion. A given rat’s sessions were separated by at least 7 days. Prior to each test, rats were habituated to the testing room for 15 minutes in their home cages. In each session, following the injection of saline/CLZ/CNO, home cage locomotion was recorded by an infrared camera for 30 minutes. Locomotor parameters including total head distance traveled and total body distance traveled were analyzed by ANY-maze (Stoelting Co).

After all locomotion experiments, spatial memory was examined using two-trial Y-maze tests (Conrad et al., 1996). Like locomotion tests, each rat was tested in all three different treatment conditions (saline/CLZ/CNO), which were randomly ordered and separated by at least 7 days.

The Y-maze was composed of three identical arms (50 × 16 × 32 cm³) made of opaque black acrylic plastic plates with a 120° angle separating neighboring arms (Conrad et al., 1996). The three arms are labeled with distinct names: Start Arm, Novel Arm, and Other Arm. There were no visual cues inside the arms. However, outside the boundaries of the Y-maze, colored paper and lab equipment with different shapes could be used as visual cues, which were rearranged across different treatment conditions (saline/CLZ/CNO) for each rat. For each subject, the order of visual cue presentation was randomized across treatment condition.

Each Y-maze test was composed of two trials. In Trial 1, an opaque black plastic plate was used to block the Novel Arm. After the injection of saline/CLZ/CNO, the rat was placed into the Start Arm and allowed to explore the Y-maze for 15 minutes. The Start Arm, Novel Arm, and Other Arm were randomly assigned for each test but remained fixed between the two trials. At the end of Trial 1, the rat was returned to its home cage outside of the testing room for a fixed intertrial interval (ITI = 2 hours). After the ITI elapsed, Trial 2 then began by placing the rat in the Start Arm; this time, though, all arms were available for exploration. The rat was allowed to explore all three arms for 5 minutes. The movement of the rat was recorded by an infrared camera fixed directly above the Y-maze. The first new arm entered (the Novel arm or the Other arm) and total time spent in each arm were analyzed by ANY-maze (Stoelting Co); 2 × 2 chi-square tests were applied to compare the number of first entries to the Novel arm between saline- and actuator-treated conditions. Percentage of total time spent in each arm was calculated for each rat and compared between saline and each actuator-treated condition using paired t tests.

In order to establish a relationship between Y-maze outcomes and RSFC data, a correlative analysis was carried out for each group (sham/unilateral/bilateral). Within each group, for a given injection (saline/CNO/CLZ), the RSFC between a given ROI pair A, B, measured by the z-transformed Pearson correlation (r) of their rsfMRI time series was averaged across an individual subject’s scans. The average correlation coefficient (⁠$r¯$⁠) produced from the saline scans was then subtracted from that of both CNO and CLZ to produce Δ$r¯$⁠. For the same subject, the Y-maze novel arm time spent in the saline condition (t) was subtracted from the novel arm time spent in the corresponding actuator condition (CNO/CLZ) to produce Δt. This process was repeated across all n subjects in the group generating a collection of values of the form (Δ$r¯$₁, Δ$r¯$₂, ⋯, Δ$r¯$_n) and (Δt₁, Δt₂, ⋯, Δt_n). Subsequently, Pearson correlation was calculated between the two lists to generate a correlation coefficient P_A,B. This process was completed across all ROI pairs, resulting in a group-specific, actuator-specific matrix of correlation coefficients P.

At the conclusion of all imaging and behavior experiments, a subset of animals was subjected to nonsurvival electrophysiology to validate viral expression functionality. Electrophysiology was recorded in four sham rats and four bilateral DREADD rats in the anesthetized condition. Rats were initially anesthetized using 3%–4% isoflurane and xylazine (12 mg/kg). The isoflurane concentration was then lowered to 0.5% and maintained at this level. In this condition, a single-channel electrode (NeuroNexus) was gradually inserted into the dHP (AP −3.5 mm, ML 2 mm, DV −2.7 mm). A reference wire was attached to the skull tissue, and a grounding wire was connected to the stereotaxic frame. Following the initiation of isoflurane administration, at least 40 minutes elapsed before the experiment proceeded; heart rate and electrophysiological signal were monitored to ensure that the animal reached a stable, lightly anesthetized state. Subsequently, the animal received an injection of saline, and the electrophysiological signal was recorded for 20 minutes as a baseline measurement. After 20 minutes of baseline recording, 0.1 mg/kg CLZ was injected intraperitoneally; electrophysiology recordings continued for another 40 minutes. The total length of recording (NeuroNexus, sampling rate = 20 kHz) was 60 min. CNO was not evaluated because appropriate responses to CLZ would necessarily imply appropriate responses to CNO (hM4Di is responsive to both actuators).

Raw data were notch filtered at 60 Hz to filter out power line noise. Multiunit activity (MUA) signal was obtained by high-pass filtering the data at 1000 Hz. Individual points of the recorded signals were classified as spikes if signal amplitude exceeded 4 times the standard deviation of the entire high-passed filtered signal, with a minimal window of 10 milliseconds following the previous spike-classified point (Tu et al., 2021).

Histology was carried out following the completion of all experiments in all animals to confirm the location of DREADD protein expression (Supporting Information Figure S1). Animals were first anesthetized and then perfused with saline and 4% paraformaldehyde (PFA) solution. The brains were stored in 4% PFA and 30% sucrose solution for a week and cut into 60-μm thick slices. Fluorescence signal was imaged using a Leica SP8 DIVE multiphoton microscope.

AAVs expressing inhibitory G-protein coupled hM4Di receptors with a pan-neuronal synapsin promoter (AAV8-hSyn-hM4Di-mCherry) were stereotactically injected into the bilateral (bilateral DREADD group, n = 15) or unilateral (right hemisphere, unilateral DREADD group, n = 13) dHP. Figure 1 shows the DREADD protein expressed in regions that coincide with dorsal DG and CA3, prominent subregions of the dHP. Control null virus vectors (AAV8-hSyn-EGFP) were injected into the bilateral dHP in sham animals (n = 8). Animals were given at least 4 weeks for recovery and protein expression. Two DREADD actuators, CLZ (0.1 mg/kg) and CNO (1 mg/kg), were used in different sessions for fMRI and behavior experiments to account for potential off-target effects, as suggested by a previous study (Jendryka et al., 2019).

To validate the neuronal suppression effect of hM4Di-DREADDs, we recorded the electrophysiological signal in the dHP before and after actuator administration in a lightly anesthetized state (isoflurane 0.5%, sham rats, n = 4; bilateral DREADD rats, n = 4). After the animal reached a stable anesthetized state, the dHP electrophysiological signal was first recorded for 20 minutes as a baseline measurement. Subsequently, 0.1 mg/kg CLZ was injected intraperitoneally as the electrophysiology recording continued for another 40 minutes. Raw data of neural recording show that the firing rate was markedly reduced after CLZ administration in DREADD rats (Figure 2A). Two-way mixed-effect ANOVAs with rat group (DREADD vs. sham) and time (baseline vs. post-CLZ window) as between and within factors, respectively, revealed that the firing rate was significantly reduced from 10 minutes after CLZ injection in DREADD animals (10–20 min window, p_interaction = 0.015) and the suppression effect persisted throughout the recording session (20–30 min, p_interaction = 0.0055; 30–40 min, p_interaction = 0.0042) (Figure 2B). In contrast, sham rats did not show significant changes in the firing rate (paired t tests: baseline vs. 0–10 min, p = 0.70; baseline vs. 10–20 min, 0.68; baseline vs. 20–30 min, p = 0.40; baseline vs. 30–40 min, p = 0.28, Figure 2B). These data confirmed the neuronal suppression effect of hM4Di-DREADDs in the dHP and null effect of CLZ alone in sham rats.

The suppression of dHP activity by hM4Di-DREADDs was also confirmed based on the rsfMRI data. We first confirmed the actuators do not have biased influence on motion or respiration rates, as these factors could alter the rsfMRI signal (Supporting Information Figure S2) (Power et al., 2014; Teichert et al., 2010). Subsequently, the activity of the dHP was quantified by fALFF, which measures the amplitude of spontaneous BOLD activity by calculating the ratio of signal amplitude in the low-frequency (0.01–0.08 Hz) range to the total signal amplitude in the full spectrum (Zou et al., 2008). Figure 2C shows the raw voxel-wise fALFF values in sham, unilateral, and bilateral DREADD rats after the administration of saline, CLZ, or CNO. In Figure 2D, voxel-wise fALFF values were averaged throughout the dHP, and the effect of the actuators was assessed by LME models. In sham rats, no statistically significant difference of fALFF in the dHP was observed between saline and either actuator-treated condition (t tests: SAL vs. CLZ, p = 0.62; SAL vs. CNO, p = 0.60; Figure 2D). Both CLZ and CNO treatments significantly decreased the fALFF in the dHP of unilateral and bilateral DREADD animals, as confirmed by two-way ANOVAs with virus type and treatment as main factors (unilateral DREADD rats: CLZ treatment, p_interaction = 4.09 × 10⁻⁴; CNO treatment, p_interaction = 0.0094; bilateral DREADD rats: CLZ treatment, p_interaction = 0.012; CNO treatment, p_interaction = 0.037. Post hoc t tests within each rat group: unilateral DREADD rats, SAL vs. CLZ, p = 1.6 × 10⁻⁹; SAL vs. CNO, p = 2.4 × 10⁻⁸; bilateral DREADD rats, SAL vs. CLZ, p = 6.2 × 10⁻⁶; SAL vs. CNO, p = 3.4 × 10⁻⁵; Figure 2D).

After confirming the suppression effects of DREADDs on local brain activities, we further assessed the impact of DREADDs on brain-wide connectivity. Consistent with the result of no significant fALFF suppression effect in the dHP in sham rats (Figure 2D), we did not observe any significant RSFC change between saline- and either actuator-treated condition in sham rats (Figure 3A, threshold: p < 0.05, FDR corrected), and the subthreshold off-target effects from the two actuators are similar (entry-to-entry Pearson correlation of their t matrices: r = 0.55, p = 1.9 × 10⁻¹⁵⁶). To control for subthreshold off-target effects of CLZ and CNO, we created sham-derived RSFC difference matrices between saline and CLZ/CNO-treated conditions. These maps were then respectively subtracted from both DREADD groups’ CLZ and CNO RSFC matrices. The resulting corrected DREADD RSFC values were compared against their corresponding saline RSFC values using LME models.

Both CLZ and CNO elicited widely spread RSFC changes in both unilateral and bilateral DREADD groups (Figure 3B). To assess cross-animal consistency of DREADD-induced RSFC changes, for each actuator we calculated the entry-to-entry Pearson correlation between the DREADD-induced RSFC difference matrix for each individual animal and the RSFC difference matrix averaged across all animals within the same group. In addition, given that both CLZ and CNO can be used to activate DREADDs, but through different pharmacokinetic mechanisms (Gomez et al., 2017), in DREADD animals, we first examined the potential difference in the effects on brain-wide RSFC between the two actuators. Our data show that within each DREADD group, the overall effects of the two actuators on RSFC matrices are consistent, reflected by significant entry-to-entry Pearson correlations of their corresponding t-value matrices (unilateral: r = 0.58, p = 4.71 × 10⁻¹⁷⁵; bilateral: r = 0.49, p = 2.08 × 10⁻¹²⁰) (Figure 3B). This conclusion was replicated even when off-target effects of CLZ and CNO were not removed from DREADD groups (Supporting Information Figure S3).

To gain a global view of the impact of DREADD suppression of dHP on RSFC, we divided the entire brain into eight systems: the hippocampus, retrohippocampus, thalamus, hypothalamus, basal ganglia, polymodal association cortex, sensory-motor cortex, and midbrain. We then assessed RSFC changes within and between these systems. Our data revealed several prominent patterns of RSFC changes that are consistent between the unilateral and bilateral DREADD groups. First, DREADD suppression of dHP dampened the RSFC within the hippocampus system in both unilateral and bilateral DREADD groups, and this effect was more pronounced in the CLZ-treated condition (Figure 3B). Second, both actuators strongly suppressed RSFC between the basal ganglia and cortex in both DREADD groups (Figure 3B). Lastly, DREADD suppression of dHP also reduced the thalamic connectivity with basal ganglia (Figure 3B). Taken together, our data indicate that dHP suppression produces a global impact on an extended hippocampal network comprised of the hippocampus, basal ganglia, thalamus, and cortex. To further validate these results, we ran a group ICA on the rsfMRI data in sham rats. Two ICA components display clear involvement of hippocampus, basal ganglia, thalamus and cortex (Supporting Information Figure S4), suggesting these brain areas indeed belong to an extended brain network.

We next examined the effect of suppressing dHP activity on the animal’s memory performance at the behavioral level. We first asked whether any of the actuators affects locomotor activity, because it could influence the exploration behavior in Y-maze tests. At the same doses as in the fMRI experiment (CLZ, 0.1 mg/kg; CNO, 1 mg/kg), the locomotion of sham rats was not affected by any of the actuators (Supporting Information Figure S5). Likewise, DREADD rats did not display any difference in home cage locomotion between saline- and actuator-treated conditions (Supporting Information Figure S5). These data demonstrate that neither actuator affects locomotion in rats regardless of the presence of the hM4Di receptor.

The spatial recognition memory of the rat was measured using a two-trial Y-maze task (depicted in Figure 4A) by recording the choice of first entry and percentage of time spent in the Novel Arm during Trial 2. Animals with intact spatial memory would exhibit preference to the Novel Arm to the Other Arm (Conrad et al.,1996). Across all treatment conditions, all sham rats exhibited strong preference to the Novel Arm in all conditions, indicated by both measures. First, in the choice of the first-arm entry, a significant bias toward the Novel Arm was observed, as indicated by the results of the chi-square tests: for saline treatment, χ²(1, n = 8) = 4.5, p = 0.034. Compared with the saline condition, actuator-treated sham rats did not show significant difference in the choice of the first-arm entry: SAL vs. CLZ, χ²(1, n = 16) = 1.07, p = 0.30; SAL vs. CNO, χ²(1, n = 16) = 1.07, p = 0.30 (Figure 4B, left). Second, sham rats with saline treatment spent significant more time in the Novel Arm compared to a chance level (i.e., 33.3% of total duration, a one-sample t test, t(7) = 2.59, p = 0.036), and this measure remains similar in both actuator-treated conditions (paired t tests; SAL vs. CLZ, t(7) = 1.08, p = 0.30; SAL vs. CNO, t(7) = 0.11, p = 0.92, Figure 4C). Taken together, these results indicate that the spatial memory of sham animals remained intact when treated with actuators alone.

For DREADD rats treated with saline, the spatial memory also remained intact (first entry: chi-square test, unilateral: χ²(1, n = 13) = 9.3, p = 0.0023; bilateral: χ²(1, n = 15) = 11.3, p = 7.9 × 10⁻⁴; time spent in the Novel Arm: one-sample t tests, unilateral: t(12) = 1.93, p = 0.079; bilateral: t(14) = 3.16, p = 0.0069). For unilateral DREADD rats, after CLZ or CNO injection, the distribution of the first-arm entry was not significantly different from that of saline injection (2-by-2 chi-square test: SAL vs. CLZ, χ²(1, n = 26) = 2.23, p = 0.14; SAL vs. CNO, χ²(1, n = 26) = 0.38, p = 0.54). Consistently, the time spent in the Novel Arm after actuator injection was also indifferent from that after saline injection (paired t tests: SAL vs. CLZ, p = 0.13; SAL vs. CNO, p = 0.16) (Figure 4B and C, Supporting Information Figure S6), indicating DREADD-induced suppression of dHP in one hemisphere is not sufficient to cause significant spatial memory loss. In contrast, for bilateral DREADD rats, both actuators caused significant loss of memory in the Y-maze test, as evidenced by the lack of preference for the Novel arm during their first entry: eight rats entered the Novel Arm, and seven rats entered the Other Arm under both CLZ and CNO treatments. This distribution significantly differed from that observed with saline treatment (2-by-2 chi-square test: SAL vs. CLZ, χ²(1, n = 30) = 6.14, p = 0.013; SAL vs. CNO, χ²(1, n = 30) = 6.14, p = 0.013). Additionally, both actuators caused significant reduction of the time spent in the Novel Arm relative to the saline treatment (paired t tests: CLZ vs. SAL, t(14) = −2.89, p = 0.007; CNO vs. SAL, t(14) = −3.46, p = 0.002) (Figure 4B and C, Supporting Information Figure S6). These results collectively demonstrate that bilateral suppression of the dHP impairs spatial memory in rats.

Next, we tested the hypothesis that changes in memory-related behavior are related to RSFC changes in brain networks. To do this, for each treatment condition RSFC matrices from repeated scans were first averaged for each animal. Then, for each actuator we calculated the correlation between actuator-induced RSFC changes for each connection in the whole-brain network and changes in time spent in the Novel Arm across all animals within the same group. This correlation analysis was conducted across all three groups of animals, comparing how each connection's RSFC change corresponded to the behavioral change in response to each actuator treatment. Figure 5 shows behavior-RSFC correlation matrices in CLZ- and CNO-treated conditions in all three groups of animals. In sham rats, these behavior-RSFC correlations derived from CLZ- and CNO-treated conditions are weakly anticorrelated (r = −0.13, p = 6.19 ×10⁻⁹) (Figure 5, upper panel). For unilateral DREADD rats, the two behavior-RSFC correlation matrices from CLZ- and CNO-treated conditions are uncorrelated (r = 0.02, p = 0.46) (Figure 5, middle panel). In contrast, the behavior-RSFC correlations are highly correlated in CLZ- and CNO-treated conditions for the bilateral DREADD group (r = 0.65, p = 1.38 × 10⁻²³⁵) (Figure 5, lower panel), indicating CLZ- and CNO-generated memory deficits share a similar underlying brain connectivity change architecture.

Considering the robust behavioral effect (Figure 4) and consistent behavior-RSFC correlations between the two actuators in bilateral DREADD animals, we further explored the network-level interpretation of impaired memory in this group. For this purpose, we constructed the brain network associated with the Y-maze memory performance by selecting connections exhibiting significant RSFC-behavior correlations under both CLZ and CNO-treated conditions (p < 0.05 for each condition). This thresholding process generated a network comprised of 45 connections. Given that all of these connections showed consistent trends in RSFC-behavior correlations between the two actuator-treated conditions in the bilateral DREADD group (Supporting Information Figure S7), we calculated one aggregated correlation coefficient between RSFC changes and behavior change (time spent in the novel arm) by pooling these data from both CNO- and CLZ-treated conditions for all individual 45 connections, as shown in Figure 6. In this network model, RSFC changes between the sensory-motor cortex (predominantly confined to the visual cortex (VIS)) and hippocampus, as well as thalamus, positively correlate with the memory performance change in bilateral DREADD rats, indicating that decreased RSFC is associated with poorer memory performance. RSFC changes between the thalamus and hypothalamus, including anterior dorsal/ventral thalamus nuclei (AD and AV) and the mammillary complex (MAM), also positively correlate with memory performance changes. In addition, RSFC changes within the midbrain show positive correlations with the memory change. Conversely, multiple thalamic nuclei display negative correlation with memory performance through their RSFC with the retrohippocampus, polymodal association cortex, basal ganglia and sensory-motor cortex (i.e., stronger RSFC is associated with worse memory performance). Taken together, our data provide a comprehensive brain network model that is linked to memory performance alteration, and support the notion that the memory function involves a properly coordinated network spanning the entire brain.

Studying the brain-behavior relationship is a central theme in neuroscience. Conventional methods use focal stimulation (e.g., electrical stimulation) or lesion to investigate how a brain region is involved in certain behavior. Although situationally useful, these historical approaches are notoriously unspecific. More modern neuroscience approaches such as optogenetics and chemogenetics allow for more spatially precise brain manipulations with cell-type specificity, which enable scientists to intentionally explore the role of specific neural circuits in behavior. However, these approaches alone are not sufficient for the investigation of large-scale brain networks, in which coordinated activity is key to mediating complex behaviors (Liégeois et al., 2019; Yamashita et al., 2015). To tackle this issue, here we incorporated chemogenetic manipulation, rsfMRI, and behavioral tests to study how changes in memory performance are caused by targeted perturbation of an extended hippocampal network in a rodent model. Our data reveal that suppressing dHP activity has a significant impact on local and global connectivity in a large-scale brain network, along with the memory performance. Interestingly, neural connections associated with the memory behavioral changes are not confined to the hippocampal formation itself but are widely distributed across other, more distant brain systems such as the cortex, thalamus, and hypothalamus. This result suggests that the impact of dHP inhibition on memory behavior is mediated by complex interactions between multiple brain systems. Additionally, we showed that two different DREADD actuators, CLZ and CNO, induced consistent effects on both brain networks and behavior tests, despite having distinct metabolic processes in the body (Gomez et al., 2017; Manvich et al., 2018) and different off-target binding profiles in the brain (Jendryka et al., 2019). These data highlight the utility of the DREADD-fMRI technology and its application in studying brain-behavior relationship.

The brain network possesses small-world properties: some brain regions have more extensive connections and play a central role in neuronal communication and integration (Oldham & Fornito, 2019). Thus, inhibiting the dHP, a central node in the network for memory processing (Geib et al., 2017), is expected to have an overall impact on brain-wide networks and memory-related behavior. Our findings are consistent with this theoretical expectation, as we observed significant network alterations not only within the hippocampal formation but also in numerous areas essential for learning and memory (Figure 3B). As a result, these network/circuit-level changes can contribute to altered performances in the spatial memory test in animals.

To further investigate this issue, we first hypothesize that DREADD-induced connectivity change between the dHP and visual cortex influences the spatial memory performance. Anatomically, the dHP receives inputs from all sensory cortices via the EC and subsequently processes the information in neural ensembles through the trisynaptic pathway of EC-DG-CA3-CA1 (Amaral et al., 2007). During visual field mapping, the hippocampus establishes a spatial representation for the visual cortex (Hindy et al., 2016; Silson et al., 2021), and the interplay between the hippocampus and visual cortex facilitates memory consolidation during slow-wave sleep (Rothschild et al., 2017; Yu et al., 2023). In addition, it has been shown that impaired hippocampal-visual cortex connectivity is related to compromised spatial representation precision in hippocampus (Domenico et al., 2021). In the current study, the long-lasting effects of actuators can compromise the memory consolidation process during the 2-hour ITI in the Y-maze test. Consistent with this hypothesis, our data show that deactivation of the dHP markedly influences its RSFC with the visual cortex, and this RSFC change is significantly associated with changes in spatial memory performance (Figure 6).

We also hypothesize that the RSFC between nodes in the Papez circuit is essential for spatial memory performance. The Papez circuit is a critical brain network involved in memory process, comprising the hippocampal formation, MAM, anterior thalamic nuclei (ATN), and retrosplenium (RSP). Anatomically, the ATN receives its main ascending projection from the subcortical MAM, which in turn receives input from the hippocampal formation. The ATN sends efferent projections to the RSP, which projects back to the hippocampal formation, completing the Papez circuit (Aggleton et al., 2016; Jones & Witter, 2007). Our data support this hypothesis, showing that changes in RSFC involving the MAM and AD/AV are significantly associated with altered spatial memory performance. These results are well in line with previous literature findings. The involvement of ATN in spatial memory has been well documented (Perry & Mitchell, 2019). Lesions to the ATN are associated with impair performance in spatial memory in rodents (Lopez et al., 2009). A recent human fMRI study also showed that the ATN is activated during visual information encoding (Sweeney-Reed et al., 2021). Besides, we found that changes in spatial memory performance are positively correlated with RSFC changes between the visual cortex and the AD and AV. Although the visual cortex is not considered part of the Papez circuit, this correlation might be mediated by other nodes within the memory-related network.

Moreover, we identified the midline thalamic nuclei (MID) as a participating node in the memory-related network, consistent with the literature that the nucleus reuniens (RE), a part of MID, receives input from the MAM and serves as a relaying center for memory information (Mathiasen et al., 2020; McKenna & Vertes, 2004). Other nodes found to be involved in the spatial memory-associated network such as the ventral tegmental area (VTA) are also in agreement with the literature report (Vann, 2009).

In addition to testing specific hypotheses, the whole-brain coverage provided by fMRI makes the DREADD-fMRI-behavior approach a valuable tool for identifying circuits that may have previously been overlooked. In the current study we discovered that spatial memory performance is positively associated with RSFC between visual and odor processing regions. Specifically, this includes previously unreported connections between the VIS and the piriform area (PIR), as well as the ventral temporal association areas (TEV) and taenia tecta (TT). The VIS and TEV are critical for visual processing (Kravitz et al., 2013), while the TT and PIR are involved in processing olfactory information (Wilson & Stevenson, 2003). Efficient communication between these sensory systems likely creates a more comprehensive representation of the environment, facilitating better encoding and retrieval of spatial information. In our experiment, although the Y-maze apparatus was cleaned with alcohol after each use, some residual odors may have remained detectable to the rats. These findings suggest that the integration of sensory information might enhance spatial memory.

Collectively, our results present an extensive memory-associated network that extends well beyond the hippocampal formation itself and underscores the importance of coordinated activity within this network in memory process. This finding highlights the power of whole-brain analysis in studying the neural mechanisms underlying complex behaviors. This advantage has also been demonstrated in previous studies. For instance, Tu et al. (2021) combined fMRI and chemogenetics with behavioral tests and found that RSFC within the default mode network (DMN) is linked to restfulness behavior, illustrating how the whole-brain approach can uncover critical functional connections for understanding behavior. Similarly, Yamashita et al. (2015) highlighted the role of the frontal-parietal network in task execution, further emphasizing the efficacy of whole-brain analyses in identifying networks involved in cognitive processes. These findings showcase the potential of combining whole-brain imaging and neuromodulation tools to advance our understanding of brain function by revealing how circuits can causally contribute to behavior and task performance. However, it is important to note that not all connections exhibiting RSFC-behavior correlations in our study play a causal role in memory performance, as RSFC changes in some connections can covary with behavior due to indirect effects from other critical nodes.

In the current study, we utilized the DREADD technique to suppress brain activity by taking advantage of its reversibility and long-lasting effect. Nonetheless, the off-target effect of actuators used in DREADDs may represent a limitation of this system. CLZ is a sedative drug that treats schizophrenia by binding to several endogenous targets, including serotoninergic, dopaminergic and adrenergic receptors (Martinez et al., 2019). Similarly, CNO binds to a wide range of receptors, such as adrenergic, histaminergic, and muscarinic receptors, with relatively high affinity (Jendryka et al., 2019). To minimize the potential off-target effects, we administered low-dose CNO and CLZ (CNO, 1 mg/kg; CLZ, 0.1 mg/kg), as recommended by literature studies (Gomez et al., 2017; Jendryka et al., 2019). In addition, our results were cross-validated using both actuators (CLZ and CNO). For both actuators, we observed consistent subthreshold alterations in brain-wide RSFC in sham rats (Figure 3A). In addition, within each DREADD group, both actuators had similar impact on RSFC (Figure 3B). In the bilateral DREADD group, both actuators impaired the spatial memory of rats in a similar manner (Figure 4). Moreover, our data show that significant correlations between behavioral and RSFC changes in both actuator-treated conditions exhibit similar trends, indicating that CLZ and CNO induce comparable effects on the brain-behavior relation (Supporting Information Figure S7). Recently, Nagai et al. (2020) introduced deschloroclozapine (DCZ) as another potent DREADD actuator. DCZ is noted for its stability and reduced off-target effects. However, its pharmacokinetics indicate that the DCZ concentrations in the plasma and brain decrease significantly within 2 hours postinjection, and it is metabolized into C21 and DCZ-N-oxide. Some studies have reported equivalent or superior results with DCZ compared to other actuators (Ferrari et al., 2022; Nentwig et al., 2022; Shimizu et al., 2023). Therefore, DCZ could be a useful alternative worth considering in future research.

By combining DREADDs, fMRI, and behavioral tests, we have demonstrated that suppressing a central node in the memory-related network induces whole-brain RSFC changes, and a subset of these changes correlate with relevant behavioral changes. Based on the associated changes triggered by the same experimental manipulation, we constructed a brain network model that is linked to memory performance. These results provide a broader and more comprehensive framework for better understanding of the neural mechanism of memory function, particularly from a network perspective.

Supporting information for this article is available at https://doi.org/10.1162/netn_a_00401.

Xu Han: Formal analysis; Methodology; Visualization; Writing – original draft. Samuel R. Cramer: Investigation; Methodology; Resources; Writing – review & editing. Dennis C. Y. Chan: Investigation; Resources; Writing – review & editing. Nanyin Zhang: Conceptualization; Funding acquisition; Investigation; Supervision; Validation; Writing – review & editing.

Nanyin Zhang, National Institute of Neurological Disorders and Stroke (https://dx.doi.org/10.13039/100000065), Award ID: R01NS085200.

Chemogenetics:

A technique that uses engineered receptors and designer drugs to manipulate cell activity.

Designer receptors exclusively activated by designer drugs (DREADD):

Genetically modified G protein-coupled receptors (GPCRs) used to selectively and reversibly control neuronal activity with specific actuator drugs.

Resting-state functional magnetic resonance imaging (rsfMRI):

An fMRI method that measures spontaneous fluctuations of neural activities at the resting state.

Resting-state functional connectivity (RSFC):

Measurement of co-fluctuation of neuronal activities in different brain regions, reflecting how these regions are functionally connected at rest.

Stereotaxic surgery:

A precise surgical technique using a three-dimensional coordinate system for injecting substances into specific brain regions.

Functional magnetic resonance imaging (fMRI):

An imaging tool that measures brain activity by detecting blood oxygenation level changes resulting from neural activity changes.

Independent component analysis (ICA):

A computational method used to separate fMRI signal into components with maximal independence by identifying consistent spatial patterns.

Linear mixed effects (LME) model:

A statistical model that accounts for both fixed effects and random effects to analyze data with multiple sources of variability.

Chi-square test:

A statistical test used to determine if there is a significant association between two categorical variables.

Multi-unit activity (MUA):

Combined neural spikes of multiple neurons nearby the recording electrode.